Fix strict YAML boolean parsing

[fallintoplace]

Summary

Replace truthiness casts for YAML-backed boolean config values with strict parsers. Quoted values like "false", "no", and "0" now parse as false instead of becoming true because they are non-empty strings.

This covers the vLLM service wrappers, Piper use_cuda, VLM MCP enable_thinking, model health_check, and voicegate listening_chime. Nemotron Omni's use_bf16 path is included too because it had the same string truthiness issue.

Invalid boolean strings now raise ValueError with the config key name instead of silently picking a truthy value.

Validation

• PYTHONPATH=server-runtime:utils/xr-ai-vllm:agent-sdk/xr-ai-models:agent-mcp-servers/vlm-mcp:utils/xr-ai-voicegate:ai-services/tts/piper:utils/xr-ai-logging:tests uv run --python 3.12 --project agent-sdk --with pytest --with pytest-asyncio --with numpy --with aiohttp --with fastmcp --with pillow --with httpx --with uvicorn --with pyyaml --with loguru pytest tests/test_vllm_lifecycle.py tests/test_models_config.py tests/test_vlm_mcp.py tests/test_xr_ai_voicegate.py tests/test_piper_tts.py::test_piper_config_parses_quoted_use_cuda_false tests/test_piper_tts.py::test_piper_config_rejects_unknown_use_cuda_string
• uvx ruff check utils/xr-ai-vllm/xr_ai_vllm/_config.py utils/xr-ai-vllm/xr_ai_vllm/__init__.py ai-services/vlm-server/vlm_server/__main__.py ai-services/llm/llama_nemotron/llama_nemotron_llm_server/__main__.py ai-services/llm/nemotron3_nano/nemotron3_nano_llm_server/__main__.py ai-services/llm/nemotron_omni/nemotron_omni_llm_server/__main__.py ai-services/tts/piper/piper_tts_server/__main__.py agent-sdk/xr-ai-models/xr_ai_models/config.py agent-mcp-servers/vlm-mcp/vlm_mcp_server/__main__.py utils/xr-ai-voicegate/xr_ai_voicegate/config.py tests/test_vllm_lifecycle.py tests/test_models_config.py tests/test_vlm_mcp.py tests/test_xr_ai_voicegate.py tests/test_piper_tts.py
• git diff --check